search

ShunChengWu / 3DSSG

Other
142 stars 22 forks source link

questions about point_dim #38

Closed xz-123-new closed 1 year ago

xz-123-new commented 1 year ago

In the the scene graph prediction task with ground truth segmentation mask, only xyz coordinates of point cloud is utilized, i'm curious that is there some limitation of this problem setting that other features like rgb information and normal vectors are forbidden to use? Or just because they are not crucial for this problem? Thanks !

ShunChengWu commented 1 year ago

We did an ablation study on those two extra inputs. They provide some improvement (especially normal) but are very minor. This repo extracts rgb information in runtime, which slows down the training significantly. If you would like to use them, you can write a new dataloader.